Release 3.3.0: HAD estimator, profile_panel, dCDH by_path, SDID survey complete

[igerber]

Summary

• Bump version 3.2.0 → 3.3.0 across diff_diff/__init__.py, pyproject.toml, rust/Cargo.toml, diff_diff/guides/llms-full.txt, CITATION.cff
• Rename CHANGELOG [Unreleased] → [3.3.0] - 2026-04-25; add comparison link
• Release date: 2026-04-25

Highlights since 3.2.0

• New estimator — HeterogeneousAdoptionDiD (Phase 1a-3 + Phase 3 pretests + Phase 4.5 A continuous-dose survey + Phase 4.5 B mass-point 2SLS / event-study survey / sup-t bootstrap + Phase 4.5 C0 QUG-under-survey gate)
• New utility — profile_panel() + llms-autonomous.txt agent-facing reference guide
• New dCDH surface — by_path per-path event-study disaggregation, including n_bootstrap > 0 percentile CIs; R-parity against DIDmultiplegtDYN 2.3.3
• BR/DR schema 2.0 — new target_parameter block
• SDID bootstrap correctness — paper-faithful refit with R-default warm-start (materially changes user SE/p-value/CI on bootstrap fits; same enum, corrected semantics)
• SDID survey support across all variance methods — bootstrap (PR Restore SDID survey-bootstrap via weighted Frank-Wolfe + Rao-Wu composition #355), placebo + jackknife (PR Restore SDID survey support for placebo and jackknife variance methods #365)
• TROP backend parity — bootstrap RNG, weight-matrix normalization, observation-weights (closes finding Address code review feedback for CallawaySantAnna covariates #23)

Methodology references

• N/A for this PR — version-bump only; no source code, methodology, or Rust changes
• Methodology citations for all underlying work are in `docs/methodology/REGISTRY.md` and the section-level CHANGELOG entries

Validation

• No test changes (release-bump PR)
• All methodology + correctness validation lives in the underlying merged PRs (Add SyntheticDiD variance_method='bootstrap_refit' and coverage MC study #351, Fix TROP bootstrap SE backend divergence under fixed seed #354, Restore SDID survey-bootstrap via weighted Frank-Wolfe + Rao-Wu composition #355, dCDH: add by_path per-path event-study disaggregation #357, Fix TROP local-method backend parity: drop Rust weight normalization + Python cache-fallthrough #358, HAD Phase 4.5: survey support on continuous-dose paths #359, dCDH by_path R-parity fixtures + TestDCDHDynRParityByPath #360, HAD Phase 4.5 B: weighted mass-point 2SLS + event-study survey composition + sup-t bootstrap #363, dCDH by_path + n_bootstrap support (library-consistent percentile CI) #364, Restore SDID survey support for placebo and jackknife variance methods #365, HAD Phase 4.5 C0: QUG-under-survey decision gate #367 — see CHANGELOG `[3.3.0]` for full list)

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall assessment

✅ Looks good

Executive summary

• The actual diff is metadata-only: release header/link in CHANGELOG.md:L8, CHANGELOG.md:L1393, version/date in CITATION.cff:L10-L11, package/runtime/build versions in diff_diff/init.py:L283, pyproject.toml:L7, rust/Cargo.toml:L3, and guide metadata in diff_diff/guides/llms-full.txt:L5.
• No estimator, weighting, variance/SE, identification, or default-behavior code changed, so there is no new methodology surface to cross-check against docs/methodology/REGISTRY.md.
• Release metadata is internally consistent: all bumped surfaces read 3.3.0, and the release date matches as 2026-04-25 in both changelog and citation metadata.
• I searched the main package/docs/build surfaces and found no stale 3.2.0 references outside historical changelog content.
• No unmitigated P0/P1 issues found.

Methodology

• No findings. This PR does not modify estimator implementations, mathematical formulas, inference code, or methodology docs; it only republishes already-written release notes under a versioned header.

Code Quality

• No findings. The changed version surfaces are consistent across runtime, packaging, Rust crate metadata, and citation metadata.

Performance

• No findings. No executable code paths or hot loops changed.

Maintainability

• No findings. The changelog release header and comparison link are coherent, and the version bump is synchronized across the standard release surfaces.

Tech Debt

• No findings. This PR does not add new deferred-work items or introduce new tracked limitations.

Security

• No findings. The diff is static metadata/documentation only and does not introduce secrets, new inputs, or new execution behavior.

Documentation/Tests

• No findings. The release metadata is aligned, and docs versioning remains anchored to package __version__ through docs/conf.py:L19-L20, so bumping diff_diff/init.py:L283 keeps the docs release string in sync. No test additions are necessary for this metadata-only PR.